Watching performance on a DHCP Server

[John Hascall]

> On Wed, 2008-02-13 at 12:04 -0600, John Hascall wrote:
> >   It seems to me that without a timeout<1>, you have an avenue for a
> >   more subtle denial of service attack -- instead of having to stream
> >   so many packets at the server that you knock it over, you just have
> >   to send enough to keep the input queue from emptying.

> the queue can only be 28 packets long.

   But does every input packet, no matter how stupidly formed,
   *always* result in something being put in the (output) queue?

> also, the code for a timeout is in but commented out for conflicting
> work (the timed event system has only recently been modified to support
> sub-second scheduling).  it's a trivial matter to put it in now, and as
> i said earlier, we will do so in 4.1.0's alphas.

   Great.

John